Research map / Detection and segmentation / CNN features
EAFormer: Scene Text Segmentation with Edge-Aware Transformers
Haiyang Yu, Teng Fu, Bin Li, Xiangyang Xue
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add edge filtering (remove non-text edges) to SegFormer (ablation) | SegFormer · TextSeg | fgIoU: 84.59 -> 86.85 | Improved |
| add edge guidance without edge filtering (ablation) | SegFormer · TextSeg | fgIoU: 84.59 -> 81.03 | Hurt |
| add edge guidance on top of edge filtering (ablation) | EAFormer · TextSeg | fgIoU: 86.85 -> 88.06 | Improved |
| add edge filtering and guidance on BTS (ablation) | SegFormer · BTS | fgIoU: 84.99 -> 88.08 | Improved |
| use loss weight lambda 1.0 instead of 0.1 (ablation) | EAFormer · TextSeg | fgIoU: 84.03 -> 88.06 | Improved |
| vary lambda from 0.5 to 10.0 (ablation) | EAFormer · TextSeg | fgIoU: 87.33 to 88.06 | No clear effect |
| introduce edge guidance at third or fourth stage instead of first (ablation) | EAFormer · TextSeg | fgIoU | Hurt |
| replace lightweight text detector with pre-trained DBNet (ablation) | EAFormer · TextSeg | fgIoU: 88.06 -> 90.16 | Improved |
| train and test on re-annotated COCO_TS | EAFormer vs TextFormer · COCO_TS (re-annotated) | fgIoU: 52.73 -> 64.82 | Improved |
| add text detection head and edge modules to SegFormer | EAFormer vs TextFormer · inference cost | parameters: 85M -> 92M parameters, 0.42s -> 0.47s per image | Mixed |
Later papers that compared with this method
2 later accepted papers report EAFormer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- ST-SAM: Multimodal Scene Text Segmentation with Dense Visual and Sparse Textual Prompts via SAM (AAAI 2026 · better on 3 of 3 shared results)
- Multi-scenario Overlapping Text Segmentation with Depth Awareness (ICCV 2025 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SegFormer (Base) — SegFormer: A Topic Segmentation Model with Controllable Range of Attention (AAAI 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TextScanner: Reading Characters in Order for Robust Scene Text Recognition (AAAI 2020 · similarity 0.88)
- Mask TextSpotter v3: Segmentation Proposal Network for Robust Scene Text Spotting (ECCV 2020 · similarity 0.87)
- ViTEraser: Harnessing the Power of Vision Transformers for Scene Text Removal with SegMIM Pretraining (AAAI 2024 · similarity 0.87)
- SRFormer: Text Detection Transformer with Incorporated Segmentation and Regression (AAAI 2024 · similarity 0.86)
- LayoutFormer: Hierarchical Text Detection Towards Scene Text Understanding (CVPR 2024 · similarity 0.86)
- OTE: Exploring Accurate Scene Text Recognition Using One Token (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
